mmdet 5: Customize Runtime Settings

자신만의 optimizer를 정의하여 사용할 수 있다.

1. Define a new optimizer

Assume you want to add a optimizer named MyOptimizer, which has arguments a, b, and c. You need to create a new directory named mmdet/core/optimizer. And then implement the new optimizer in a file, e.g., in mmdet/core/optimizer/my_optimizer.py:

from .registry import OPTIMIZERS
from torch.optim import Optimizer


@OPTIMIZERS.register_module()
class MyOptimizer(Optimizer):

    def __init__(self, a, b, c)

2. Add the optimizer to registry

To find the above module defined above, this module should be imported into the main namespace at first. There are two options to achieve it.

• Modify mmdet/core/optimizer/__init__.py to import it.The newly defined module should be imported in mmdet/core/optimizer/__init__.py so that the registry will find the new module and add it:

from .my_optimizer import MyOptimizer

• Use custom_imports in the config to manually import it

custom_imports = dict(imports=['mmdet.core.optimizer.my_optimizer'], allow_failed_imports=False)

The module mmdet.core.optimizer.my_optimizer will be imported at the beginning of the program and the class MyOptimizer is then automatically registered. Note that only the package containing the class MyOptimizer should be imported. mmdet.core.optimizer.my_optimizer.MyOptimizer cannot be imported directly.

Actually users can use a totally different file directory structure using this importing method, as long as the module root can be located in PYTHONPATH.

3. Specify the optimizer in the config file

Then you can use MyOptimizer in optimizer field of config files. In the configs, the optimizers are defined by the field optimizer like the following:

optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)

To use your own optimizer, the field can be changed to

optimizer = dict(type='MyOptimizer', a=a_value, b=b_value, c=c_value)

Some models may have some parameter-specific settings for optimization, e.g. weight decay for BatchNorm layers. The users can do those fine-grained parameter tuning through customizing optimizer constructor.

from mmcv.utils import build_from_cfg

from mmcv.runner.optimizer import OPTIMIZER_BUILDERS, OPTIMIZERS
from mmdet.utils import get_root_logger
from .my_optimizer import MyOptimizer


@OPTIMIZER_BUILDERS.register_module()
class MyOptimizerConstructor(object):

    def __init__(self, optimizer_cfg, paramwise_cfg=None):

    def __call__(self, model):

        return my_optimizer

The default optimizer constructor is implemented here, which could also serve as a template for new optimizer constructor.

Optimizer로 실행할 수 없는 트릭들은 optimizer constructor 혹은 hook을 통해 실행한다. 다음은 학습을 안정화 혹은 가속화시킬 수 있는 몇 가지 일반적인 세팅들이다.

Tricks not implemented by the optimizer should be implemented through optimizer constructor (e.g., set parameter-wise learning rates) or hooks. We list some common settings that could stabilize the training or accelerate the training. Feel free to create PR, issue for more settings.

• Use gradient clip to stabilize training: Some models need gradient clip to clip the gradients to stabilize the training process. An example is as below:optimizer_config = dict( _delete_=True, grad_clip=dict(max_norm=35, norm_type=2)) If your config inherits the base config which already sets the optimizer_config, you might need _delete_=True to override the unnecessary settings. See the config documentation for more details.

• Use momentum schedule to accelerate model convergence: We support momentum scheduler to modify model’s momentum according to learning rate, which could make the model converge in a faster way. Momentum scheduler is usually used with LR scheduler, for example, the following config is used in 3D detection to accelerate convergence. For more details, please refer to the implementation of CyclicLrUpdater and CyclicMomentumUpdater.